Está en la página 1de 11

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
1

An Internet of Things Framework for Smart


Energy in Buildings: Designs, Prototype, and
Experiments
Jianli Pan*, Raj Jain‡, Subharthi Paul◊, Tam Vu§, Abusayeed Saifullah#, Mo Sha※
*
University of Missouri – St. Louis, pan@umsl.edu
‡ ◊
Washington University in St. Louis, Cisco Systems, Inc.
§ #
University of Colorado – Denver, Missouri University of Sci. & Tech.

Division of Science, Governors State University

 negative impacts on the planet’s climate and ecological balance


Abstract— Smart energy in buildings is an important research make it important for us to explore new clean-energy sources
area of Internet of Things (IoT). Buildings as important parts of and improve the energy efficiency in the consumer-side smart
the smart grids, their energy efficiency is vital for the environment grids of various buildings.
and global sustainability. Using a LEED-gold-certificated green
However, buildings are complex systems and many factors
office building, we built a unique IoT experimental testbed for our
energy efficiency and building intelligence research. We first can affect the total energy consumption in different buildings.
monitor and collect one-year-long building energy usage data and Also, conventional buildings are not with too many intelligent
then systematically evaluate and analyze them. The results show designs. It is meaningful to monitor the real energy
that due to the centralized and static building controls, the actual consumption data and find the major factors and patterns
running of green buildings may not be energy efficient even through systematic modeling and analysis for different types of
though they may be "green" by design. Inspired by "energy
buildings. Such results can be used to further design and
proportional computing" in modern computers, we propose a IoT
framework with smart location-based automated and networked implement appropriate IoT based networking system to
energy control, which uses smartphone platform and cloud construct appropriate methods and strategies improving the
computing technologies to enable multi-scale energy energy efficiency for both “green” and “non-green”
proportionality including building-, user-, and (conventional) buildings. We summarize the research on the
organizational-level energy proportionality. We further build a topic into three sequential key aspects:
proof-of-concept IoT network and control system prototype and
(1) Energy Monitoring: Through communication networks,
carried out real-world experiments which demonstrate the
effectiveness of the proposed solution. We envision that the broad the consumption and generation of energy are monitored and
application of the proposed solution has not only led to significant logged in different granularities including the whole building,
economic benefits in term of energy saving, improving floors, departments, labs, rooms, and even occupants.
home/office network intelligence, but also bought in a huge social (2) Energy Modeling and Evaluation: Through off-line
implication in terms of global sustainability. modeling and evaluation, identify the energy consumption
patterns and factors that may influence the consumption and the
Index Terms— Internet of Things; Smart Energy; Energy
Efficiency; Multi-scale Energy-Proportionality; Intelligent extent of their impact.
Buildings; Location-based Networked Control (3) IoT System to Apply Practical Changes and Strategy
Adjustments: The modeling and evaluation results are used to
I. INTRODUCTION identify the key energy components of the building, to apply
adjustments, and to devise strategies to reduce energy
Smart energy in buildings is an important research area of
consumption. IoT based networking system is designed and
Internet of Things (IoT). Buildings as important parts of the
prototyped to realize the strategies and achieve the goal.
smart grids, their energy efficiency is vital for the environment
Our research covers all the three aspects. We monitored and
and global sustainability. According to a general survey [1], in
United States, buildings are responsible for around 38% of the collected the building energy usage data for almost a year.
Based on our data traces, we systematically identified the
total carbon dioxide emissions; 71% of the total electrical
energy consumption patterns and explored potential methods to
energy consumption; 39% of the total energy usage; 12% of
water consumption; 40% of non-industrial waste. In the improve the energy efficiency. The results show that due to
centralized and fixed pattern control, the actual running of
meantime, cost of traditional fossil fuels is rising and its
green buildings may not be energy efficient even though they
may be "green" by design. Inspired by "energy proportional
Copyright (c) 2015 IEEE. Personal use of this material is permitted. computing" in modern computers, we propose a smart
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org.
location-based automated energy control IoT framework using

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
2

smartphone platform and cloud computing technologies to II. ENERGY EFFICIENCY EVALUATION AND ANALYSIS
enable smart mobile control and multi-scale energy In this section, we describe our energy monitoring testbed,
proportionality, which includes building-, user-, and evaluation methodology, and data modeling and analysis
organizational-level energy proportionality. We further built results. This section is mostly based on our contributions in [3].
an experimental IoT prototype system to demonstrate the
effectiveness of our proposed idea. Our results show potential A. Energy Monitoring Testbed and Justification
economic and social sustainability benefits. We realize that buildings can be very different from each
Unlike simulation based solution, our work is based on real other and it is extremely important to find the common “thing”
measured data traces for a currently in-use on-campus green or pattern among them in terms of energy efficiency. So in our
building, and a real IoT system to control the energy project we talked to some on-campus building maintenance
automation. We use the latest information technologies such as experts thoroughly and investigated the common structure of
mobile smartphones with location service, distributed control, these buildings in U.S. For our testbed, we pick a very typical
office building constructed in 2010 (actually two latest
and cloud computing to actively involve the occupants in the
on-campus buildings constructed in 2014 also show that they
energy-saving process. Energy-saving policies from multiple
share exactly the same technologies as our testbed due to the
sources such as individuals and organizations are considered in reason that traditional building-based systems are not an area
an integrated policy framework in deciding the final energy that is developing as fast as those IT technologies). Besides our
saving strategies. We aim to create an energy-efficiency IoT experimentation on this particular office building, we also
testbed that can be easily migrated to all kinds of buildings and investigated a case in another extreme end, a Net-Zero Energy
achieve energy savings in multiple scales. Building named “Tyson Research Center”, which is a small
In this journal version paper, we first summarize and refine office building. It received prestigious award in the “living
our previous work in two conference papers [2, 3]. In [2], we building challenge”. We found that for such small office
evaluated the building energy usage data and presented our buildings or home buildings, it is relatively simpler and easier
findings in identifying the major issues in these buildings. to apply networking technologies to control or change their
Based on that, we proposed a smart location-based networked energy policy. In comparison, large buildings like our testbed
energy control IoT system design to tackle the issue and are more difficult to change and it is also one of the reasons
improve the energy efficiency [3]. why in this paper we primarily focus on such large office
In this paper, however, we add new contributions to buildings. With our findings in this testbed, it is relatively
complete the three steps described above. Particularly, easier to tailor and generalize the system to solve the issue with
(1) We synthesize the previous separate contributions into a other buildings of the same type or different types.
complete IoT framework design. It includes research and Our testbed building received a Gold certificate from LEED
work in the whole process of identifying the key problems, rating system [4] by U.S. Green Building Council (USGBC) [5].
finding methods to solve them, and developing prototype It adopts a series of energy efficiency and sustainability
features. The overall resource usage data for the building are
system to prove the effectiveness of the proposed method.
monitored and logged through a series of meters every 30
(2) We build a novel experimental prototype IoT system which
minutes (some are 15 minutes) through wired network for
demonstrates the real time location-based automated energy future off-line data modeling and analysis. It is a very typical
policy control across multiple buildings. It is the basic step large green office building with typical monitored subsystems
in changing from the current centralized control and static such as HVAC (Heating, Ventilation, Air-conditioning, and
energy consumption modes to distributed and dynamic Cooling), lighting, and water systems. We believe that the
energy control in the consumer-side smart grids containing experiments and further data analysis findings from this testbed
various common buildings. apply to other large office buildings.
(3) Based on these, we propose to create a future of multi-scale
energy proportionality. The central idea is to generalize the B. Data Source and Analysis Methodology
smartphone and location-based energy control idea and We sorted out the most useful measured data by analyzing
include policies of multiple levels of organizations. It the relationships among various parameters. Based on it, the
aggregates the energy saving of individual users and allows data points that we use include: the total electrical energy
distributed and dynamic energy control, which is the key consumption, the heating and cooling energy consumption, and
for energy proportionality. the outdoor and indoor environmental data such as
temperature and humidity. The heating and cooling parts are
The rest of this paper is organized as follows. Section II deemed as the HVAC consumption while the total electricity
presents the testbed description, methodology, and detailed consumption covers a wider range of loads in the building.
energy efficiency data analysis and discussions. Section III Though separate lighting data may be useful, such data is not
describes our idea on the smart location-based automated currently available. Moreover, we unify the semi-hourly or
energy control IoT architecture. Section IV discusses the hourly logged data to an hourly basis for uniform analysis.
prototype system and the experimental results, and the Our primary modeling and evaluation goal is to identify the
discussions on multi-scale energy proportionality. Section V energy consumption pattern and know how it is related to: (1)
reviews some related work and Section VI concludes the paper. environmental factors, and (2) occupancy rate. So, we first
analyze the relationship between electricity, heating, and

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
3

cooling energy consumption and the outdoor environmental 650

Electr. Consumed (kWh)


factors. Our method is to combine the short period (longer than 600
1 day and less than 1 week) and the long period (several months)
550
correlation analysis to show the overall trends. We group the
hourly data into multiple granularities such as weekly and 500
monthly to reveal the complete correlation differences over a 450
relatively long period. We also develop Multiple Polynomial
Regression (MPR) model and Multiple Linear Regression 400
(MLR) model to reveal longer term average seasonality trends. 50 60 70 80 90

C. Detailed Evaluation and Analysis


Temperature (degree F)
Fig. 1. Total electrical energy consumption with temperature
In this section, we present the detailed modeling and

Heating Energy (BTU)


evaluation results and the corresponding analysis. 1.6M
1) Environmental Impacts Analysis
Here, we focus on temperature and humidity, and study their 1.2M
impacts on the total electrical and HVAC energy consumption. 800k
a) Short Period Basic Trend and Correlation Analysis
We put two groups of factors together: (1) group 1 made of 400k
electric consumption, heating energy, and cooling energy; (2)
group 2 containing temperature and humidity. We want to see if 0
50 60 70 80 90
there is any straightforward connection. Fig. 1 shows the
relationship between electrical energy consumption and Temperature (degree F)
Fig. 2. Heating energy consumption with temperature
temperature. It shows almost no correlation.

Corr(ZY) and Corr(ZX)


Fig. 2 shows the relationship between heating energy and 1.0 Spring Summer Fall
temperature, in which we still do not find very strong 0.8 Winter
correlation (due to the similar patterns between heating and Corr(ZX)
cooling, in this paper, we only show results of heating data). 0.6 Corr(ZY)
Note that in the figures we use the British Thermal Unit (BTU) 0.4
as the unit for heating and cooling. 1 BTU is equal to 1055 joule
0.2
or 0.293 watt-hours.
Observation: (1) Overall, correlations between both heating 0
and electric energy consumption and the outdoor weather 0 10 20 30 40
conditions are small. (2) Overall, the electrical consumption Weeks
also shows very little variation between days and nights, which Fig.3. Correlations between electrical energy (Z), temperature (X), and
means that it possibly has a small correlation with occupancy. humidity (Y)
Corr(HY) and Corr(HX)

b) Long Period Correlation Analysis 1.0


Spring Summer Fall
We now study the correlation among multiple factors over a 0.8 Winter
Corr(HX)
longer period. After filtering out incomplete and inaccurate
data, we get a continuous dataset for about 10 months (39 0.6
weeks). It ranges from 3/18/2011 to 12/31/2011. We group the 0.4
data into weeks and every week has 24*7=168 data points. For
each 168 data point set, we calculate the correlations among 0.2
Corr(HY)
multiple factors. These factors include: temperature (denoted as 0
X), humidity (Y), total electrical energy consumption (Z), 0 10 20 30 40
heating energy (H), and cooling energy (C). We also mark the Weeks
seasons according to the Missouri climate convention. Fig. 4. Correlations between heating energy (H), temperature (X) and
humidity (Y)
The correlations between electrical energy consumption and
weather conditions are shown in Fig. 3. They are mostly below
0.5. Interestingly, the correlation for summer season is a bit c) Daily Average Data Analysis
higher than that for fall and winter seasons. The results validate So far, we studied hourly electricity, heating, and cooling
the results we showed in Fig.1 and 2. Note that the X and Y in energy data (1 data sample per hour). We also aggregate the
the figures do not mean x-axis and y-axis, but temperature and data into daily averages to see if there are any new findings.
humidity in our notation. The results for the correlations of Specifically, we calculate the daily average temperature and
heating energy with weather conditions are shown in Fig. 4. humidity, and the daily total electric, heating, and cooling
Observation: The figures roughly indicate that the heating energy consumption. Then we have a data set for each day and
and cooling systems do not actively take the outdoor weather a total 245 data sets from 5/1/2011 to 12/31/2011.
condition as factors to dynamically adjust the running The daily heating and cooling trends are shown in Fig. 5. The
schedule and policies to save energy. seasonality is clear for both heating and cooling data in that

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
4

there is more cooling and less heating energy in the summer. In numbers for after hours and weekends are not as low as
total, for the above period, the energy usage is 8.1 billion BTU expected which also illustrates that the current building
for heating and 16.9 billion BTU for cooling. It is interesting operation is far from efficient and is not proportional to the
that the cooling system uses about twice the energy than heating. actual usage or occupancy.
In the summer months the cooling energy usage is The heating energy pattern shown in Fig. 8, however, is a
significantly higher than that in other months. The daily little bit different. Overall, the heating energy consumption for
electrical energy consumption in Fig. 6 shows a very regular afterhours is about 6% higher than weekends, and 19% higher
fluctuation. Seasonality is not that obvious. than those for business hours.
Observation: the electricity provisioning in this building is

Daily Heating&Cooling (BTU)


Summer
Spring Fall Winter
relatively fixed and “extra capacity” is generally provided to 160M
satisfy any burst usage. In other words, a lot of electrical Daily Cooling
120M
energy is wasted, especially, during afterhours. Daily Heating
80M
d) Regression Modeling and Analysis
We further use regression models to analyze the 40M
relationship among multiple factors and use the statistical
0
approaches to examine whether they can justify our findings.
We try both Multiple Polynomial Regression (MPR) and 06/01 08/01 10/01 12/01
Multiple Linear Regression (MLR) models, and compare the Days
two results. Fig. 5. Daily heating and cooling
First, we use the same daily average dataset and we have a

Daily Electricity (kWh)


vector of data point for each day. The vector is <daily average 12k
temperature, daily average humidity, daily electrical energy,
daily heating energy, daily cooling energy> and we have 245 11k
data vectors in total. We compute the coefficients of each factor
in the two types of regression models, calculate the errors and 10k
conduct tests to check the models' effectiveness.
Table I presents results of the MPR and MLR on electrical, 9k Spring
Summer Fall
Winter
heating, and cooling energy predictions with temperature and
8k
humidity as two parameters. As shown in Table I, the 06/01 08/01 10/01 12/01
coefficient of determination R2 is the fraction of the total Days
variation explained by the regression [6]. For example, for Fig. 6. Daily electrical energy consumption
electrical energy MPR, R2 is 0.1902 which means that the
MPR regression model can only explain 19.02% of the Table I. Regression results for electricity, heating, and cooling energy
Electrical Energy Heating Energy Cooling Energy
variation of electrical energy usage. In comparison, the R2
MPR R2 0.1902 R2 0.8634 R2 0.9884
value for cooling energy is 0.9884 which means that the MPR 2 2
MLR R 0.0213 R 0.8610 R2 0.9072
can explain 98.84% of the variation of the cooling energy
consumption. This result validates our previous conclusion. 16K
Office hours,
Electrical Energy in

Observation: The regression reminds us that various energy weekdays


subsystems of the buildings are impacted differently by the 12K
24 hours (kWh)

environmental factors; hence, for better energy efficiency, we


should tune each subsystem separately. For example, heating 8K
and cooling respond more to the environment and we may use Whole day, After hours,
environment condition to tune the running policy of the HVAC weekends weekdays
4K
system and save energy.
2) Occupancy Impact Analysis 0
In this section, we focus on the occupancy and study how it 5/1 7/1 9/1 11/1 1/1
can impact the energy consumption. Weeks
Fig. 7. Comparison of electrical energy consumption averaged in 24 hours for
a) Weekdays/Weekends Energy Comparisons office hours, after hours, and weekends
We roughly divide the data into three subsets: regular
office hours (8:00am to 8:00pm of weekdays), after hours It is interesting to know that the heating consumption for the
(8:00pm to 8:00am of weekdays), and weekend (whole days of business hours is the lowest compared to the other two. It is
Saturday and Sunday). We study the data by weeks and for probably due to the fact that the occupancy rate is higher
every week we have three subsets. For each subset, we during the office hours and more people are active and
calculate their electrical and heating energy averaged in 24 providing body heat in the building and hence reduce
hours, and compare them to see the differences. The results are external heating energy demands.
shown in Fig. 7 and Fig. 8, respectively. From Fig. 7, we can Observation: the analysis clearly shows that the actual
see that the electrical energy consumption during office hours is occupancy rate has very low impact to the energy
about 15% more than that for afterhours and weekends. The

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
5

consumption. Ideally, the numbers for after hours and We will discuss our major idea and corresponding prototyping
weekends should be much lower than those for office hours. and experimentation in the following two sections.
50M

Daily Electrical
Heating Energy in
12K

Energy (kWh)
After hours,
24 hours (BTU) 40M weekdays Summer Fall
30M 8K

20M Office hours,


weekdays 4K
Whole day,
10M weekends
0
0 Electricity
5/1 7/1 9/1 11/1 1/1
Fig. 9. Summer and Fall daily electrical energy consumption comparison
Weeks

Cooling Energy (BTU)


Fig. 8. Comparison of heating energy consumption averaged in 24 hours for

Daily Heating and


office hours, after hours, and weekends 100M
b) In-semester/Holidays Energy Comparison 80M
To further see the occupancy rates' impact, we selected out Summer
60M Fall
the data for in-semester days and summer holidays. Based on
the academic calendar of Washington University for year 2011, 40M
we pick the period between Aug. 30 to Dec. 09 (101 days in 20M
total) as the fall semester, and the period between May 10 to
0
August 29 (111 days in total) as the summer holiday season. Heating Cooling
Generally, the summer holiday season has lower occupancy Fig. 10. Summer and Fall daily heating and cooling energy consumption
rate than the regular fall semester in our testbed. comparison
Firstly, we compute the total electrical energy consumption 100
for the above two periods and average them by the days. The Daily Energy Consumption
results are shown in Fig. 9 which indicates that the electrical Fall vs. Summer 80
consumption varies very little for these two periods. We also
compared the heating and cooling energy consumptions which 60
are shown in Fig. 10. Daily average heating energy of the fall 40 Top: Summer
season is about 20% higher than summer, while daily cooling is Bottom: Fall

65% lower than summer season. Such results are consistent 20


with the analysis results of the previous several subsections. If
0 A: After hours
we separate the two periods into office hours and after hours, A O A O A O O: Office hours
then we have a more detailed view of the energy consumption
patterns. As shown in Fig. 11, we scale the daily energy Electr. Heating Cooling
consumption in Y axis into a 0 to 100 range. We find that for Fig. 11. The daily energy consumption comparison between Summer and
Fall, considering after hours and office hours
electricity usage during after hours, it is almost fifty-fifty
between summer and fall seasons, while summer is a little bit
higher than fall for office hours (first two columns in Fig. 11). III. SMART LOCATION-BASED AUTOMATED ENERGY
Summer and fall heating energy are almost even for both after CONTROL FRAMEWORK
hours and office hours (middle two columns in Fig. 11). In this section, we present our smart location-based
Afterhours and business hours have also close cooling energy automated energy control IoT framework. This section is
consumption (the 5th and 6th columns in Fig. 11). mostly based on our contributions in [2].
Observation: the comparison in different granularities
shows that there is no direct and visible connection between A. Overall Structure
the energy consumption and the occupancy rate. In other There are multiple design components and aspects which
word, a lot of energy is wasted regardless of the actual usage. interact with each other and form a complete framework of our
To summarize our findings with the energy efficiency idea to fulfill the goals. We envision an occupant oriented and
modeling and evaluation results in the building-side smart grids involved networked system and depict it in Fig. 12.
and the HVAC system energy consumption, we find that both The key design components include: mobile devices based
green buildings like our testbed and most of the conventional distributed energy monitoring and remote control, location
buildings are with centralized control and fixed running application on smartphone, multi-source energy-saving policies
patterns, which leads to poor energy efficiency in operations and strategies, cloud computing platform based data storage
though some of them may originally be designed to be "green". and application, and energy data modeling and strategy
We further argue that to enable more energy efficient formation. We discuss these below.
consumer-side building grids, we need to find a series of
networking designs to enable distributed and dynamic control.

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
6

B. Smart Mobile IoT Devices as Remote Controls to signal strengths from various transmission towers. New
In the last several years, smart mobile devices have become generations of smartphones can provide localization much
very popular. Smartphones generally have multiple networking more precisely with embedded GPS chips. We use this location
interfaces such as 3G, WiFi, WiMAX, Bluetooth, and have information in designing automatic control policies that can
multiple sensors including GPS sensors. Because of various turn on/turn off energy consuming devices at home or office
connectivity provisions and global accessibility to the Internet, depending upon the location and direction of movement of the
they are suitable for use in any system that needs humans’ user. By doing so, a dynamic and flexible policy can be applied
online participation or interaction. The “Internet of Things” [7] which satisfies the user’s preferences for energy saving and
trend makes the cost even lower and the sensors are connected comfort. An “App” on the device can automatically enforce
to the Internet at all time. these desired policies.
Smartphones are ideal for monitoring, controlling, and Location Service
managing the energy control systems remotely from anywhere App. On
at any time. After appropriate authentication and authorization, Smart Phone
the occupants are allowed to modify and change their Distributed
Monitoring
energy-saving policies online by interacting with the policy Cloud Storage, Occupants- And Control
servers of their office and residential buildings. Such design Computing, and Involved
allows dynamic changes to the energy-saving policies and App Ext. Green-building
Testbed
offers better flexibility to the occupants. It can be a good
complement to the general policy decision process based on the Multi-source
modeling results. Such an “app” can be easily developed for the Modeling Energy-saving
On Location Policies and
smartphone based on the web technology.
Pattern strategies
C. Multi-source Energy-saving Policies Hierarchy Fig. 12. Overall structure of our design with components and their interaction
In a real environment, various parts of an organization, such
as campus, building, department, and labs may be in charge of Building Policy
Server
different components of a building. Each of these may have
their own policies and requirements that need to be taken care Dept. 1 Policy Dept. 2 Policy Dept. 3 Policy
Server Server Server
of in controlling the energy consumption. Even in a single
home building, locations of multiple family members and their Lab 1 Policy ② Lab 3 Policy
Server Server
preferences need to be taken into account. Therefore, in our Lab 2 Policy

location based automatic control scheme, we add policies Server Office control plane

coming from these levels of control hierarchy. Office building Data storage
in the cloud
Fig. 13 shows an example of the policy hierarchy. As shown,

there may be a tree-like structure for the building control plane Data centers
3G network, Internet
in which there are policy servers enforcing the energy-saving
policies covering different levels. This also applies to the Users with
residential buildings in which the tree structure may be mobile devices

Users with
relatively simple. The mobile users can be connected to the mobile devices
Internet through smartphone, tablet, or even laptop with WiFi

connections. In the example shown in Fig. 13, the mobile Home control plane

smartphone holder leaves the home building and travels
towards his office building. The movement and location Home building

changes will trigger the policy servers to adjust the Fig. 13. Example dynamic multi-source energy-saving policy adjustment by the
mobile devices
energy-saving policies for both buildings accordingly. The
action steps are denoted as “①②③” in the figure.
With the help of the location-aware mobile devices, these
In our previous research on next generation Internet [8, 9] as
well as the policy-oriented Internet architecture [10], we have dynamic adjustment policies could also enable the cooperation
and interaction among different buildings. For example, when
experimented with several policy based control schemes. We
apply similar ideas to the building and community the location detection daemon on the user’s smartphone detects
that the user has moved out of a threshold distance range from
environments. In particular, each control region can be defined
his home building and is moving into a threshold distance range
as a “realm” [11] which is managed by a realm manager (also a
policy server in our building testbed). Energy control policies of his office building, then a message is sent to a centralized
server to trigger the policy control process. The office building
may span multiple realms and sometime conflicts may have to
be resolved. room owned by the user will start pre-heating/cooling to
prepare a user-customized or optimized working environment,
D. Mobile Device Location-Based Automatic Control while the message also triggers the home building to transit into
Almost all phones can determine their location by referring an energy-saving mode.

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
7

E. Cloud Computing and Storage implement their own energy policies in real time, and enable
Cloud computing has become very promising in the last few their energy consumption to be proportional to their actual
years. We have two basic kinds of jobs which need the usage.
cloud-computing platform: (1) The cloud-based data storage, Note that in this simple prototype, we only implement the
and (2) the cloud-based modeling and analysis computation. case involving only one user with control devices in two
buildings (the user’s office building and home building). In
We have a preliminary design of how to integrate the system
other words, we implement this small-scale proof-of-concept
into the cloud computing platform. As shown in Fig. 14, the
system and compare the energy saving with the case that is
cloud provides the basic data storage and retrieval service for
without the new design. After proving the effectiveness, then
the logged building energy consumption data. we could generalize it into a larger scale. In the future work, we
Computation-intense modeling and analysis jobs are mostly plan to test the case with multiple users controlling their
done in the cloud. The communication layer provides devices simultaneously by which we could show results with a
configurability, reliability, and security for the network larger scale energy saving.
communication between the cloud and the client. The middle 1) Hardware and networking structure
layer in Fig. 14 is for cloud application development by using In the prototype system, the hardware systems that we use
the open API provided by the cloud providers such as Google include the "Kill-A-WattTM" electrical meters [14], WeMoTM
App Engine. The reason we incorporate this layer in our design control devices [15], servers in each building which act as both
is that it can alleviate the overhead to develop the cloud web daemon server and in-building controller, WiFi routers,
application and accelerate our application development and and smart devices with location sensors (GlobalSat GPS
deployment process. It also becomes much easier to integrate module).
other services using the same platform (such as authentication The networking structure of the prototype system of the
services, email services and user interfaces) to the application home building side is shown in Fig. 15. The basic function is
on demand and make the development of a cloud application a that a smart mobile device with a location sensor keeps sending
less complicated task. The top layer is the application layer. We its location data back to the web servers inside the home
are researching and developing a user-friendly prototype building and the office building. The web daemon servers
web-based user interface and application for the building behind the firewall and NAT (Network Address Translation)
are accessed from outside by port mapping technology. It also
environment, which can be easily configured and managed by
calculates the distance between it and the mobile devices to
the remote client.
decide if the distance passes a specific threshold to trigger
energy policy changes in either of the buildings. If it does, then
it initiates the controller to send instructions to turn on/off
Web-based Application specific devices in its territory according to the energy policies.
Software Platform: API
Cloud Side
storage Modeling Comm.
Kill-A-Watt WeMo
meter Control
Comm.
Monitored Results Capability
Data GPS Location in
UPnP NAT
NMEA-0183 format
Web-based app: Appliances
Application allow easy configuration
Development and management
… Internet
Home/Office
Controller/
Building Side WiFi
FTP/Web Server Port Mapping

Fig. 14. Cloud computing components and interaction with the building side Python & CGI codes
servers Home/Office Building Boundary
Fig. 15. Prototype system networking structure
IV. PROTOTYPE SYSTEM, EXPERIMENTAL RESULTS, AND 2) Software
MULTI-SCALE ENERGY PROPORTIONALITY The software part includes the software for GPS location
data recording and sending to the web server in NMEA
This section focuses on the prototype system we build to
(National Marine Electronics Association) 0183 compliant
prove the effectiveness of the idea proposed in Section III.
format, and the WiFi router's configuration and management
A. IoT Prototype Description software which provides a port mapping service for web access
In this IoT prototype system, we implemented a simple from outside of the NAT. The web server is programmed with
scenario involving a user associated to two groups of electrical CGI (Common Gateway Interface) scripts to execute Python
appliances: those in his/her home apartment and those in his/her codes controlling the WeMo devices through UPnP (Universal
office room. It is a simplified scenario of what is shown in Fig. Plug and Play) protocol. Besides the location based automated
13. Our goal is to provide users the ability to dynamically adjust control, these software parts working together with the
and control their devices across two buildings. The basic hardware also enable the devices in both buildings to be
function is to enable the server to detect the user's location controllable from Internet in real time through smart devices.
changes and trigger the energy policy changes by turning on/off B. Experiments and Results
the electrical appliances in both buildings associated to the user.
We first measure the baseline electricity appliances' power
By doing this, we essentially enable users to control and
associated with the user in both buildings. The major

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
8

appliances in the home building of the prototype system and and laptop, whose on/off status do not directly affect the normal
their baseline power measurement and estimation are shown in living of the human being.
Table II. Note that in this prototype system, we primarily focus
Table V. Daily office electricity consumption estimation of three modes
on electricity appliances, though in the real case, HVAC can be
Lighting Desktop Laptop HVAC
a significant energy consuming source worth applying dynamic
control to make a difference in improving energy efficiency. Luxury Mode Always ON 24/7 Always ON when at
Similarly, the appliances in office room and its baseline power (user is energy office
insensitive) 160W*24 50W*24*1/3=
measurements are shown in Table III. =3.84kWh 0.4kWh
To compare and quantify the real savings of our prototype Moderate Always ON when at Only ON when at Only 50% ON when
system, we divide the users' energy usage into three potential Mode office office at office
192W*8= 1.54kWh N/A
modes: luxury mode, moderate mode, and frugal mode. For 160W*24*1/3 0.4*0.5= 0.2 kWh
=1.28kWh
each mode, we estimate how much energy will be consumed on Frugal Mode Only 60% ON OFF when at office,
a daily basis. The estimation results for home and office are (user is energy when at office use desktop
sensitive)
shown in Table IV and Table V respectively, which also 1.28*0.6=0.77kWh 0kWh

explain the three modes. Total Luxury: 5.78kWh ; Moderate: 3.02kWh; Frugal: 2.31kWh

*Assuming 8 hours working in office, 8 hours at home awake, and 8 hours sleeping
Table II. Home electricity appliances' baseline power measurements
Type Lighting Refrigerator Microwave Laptop (Mac HVAC
(GE) Stove(Philips) Pro 15’’)
Items Porch: 54W Start: 200W, 1.3kW Normal: 41W N/A
Bedroom: 18*2 = 36W gradually to
170W Active or
Living 54*2+42 = charging:
Room: 150W Compressor 60W
Kitchen: 52*5 = 260W work for 9 min,
Bathroom: 54W stop for 9 min

Avg. Power 550W 185/2 W 1.3kW 50W

Table III. Office room electricity appliances' baseline power measurements


Type Lighting Desktop Laptop (Mac HVAC Fig. 16. Location history in a 24 hours' period
Pro 15’’)
Items 32W * 6 = Host: Normal: 41W N/A In the real activity trace of our experiments, it shows that
192W Boot – 110W
Normal – 67W Active or
approximately the user spent 14 hours at home in which 8 hours
charging: for sleep, 2 hours for lunch and rest, and 4 hours for working at
Monitor:
60W home. The total recorded real energy consumption at home
Normal—72W, Active—
80~90W during this period is 5.285 kWh, which includes 2.7kWh for
Avg. Power 192W 160W 50W lighting, 2.22kWh for refrigerator, 0.065kWh for microwave
stove, 0.3kWh for laptop.
Table IV. Daily home electricity consumption estimation of three modes Also, during this period, the appliances in office room are
Lighting Refrig. Microwave Laptop HVAC kept in "OFF" status by the control server of the prototype
Luxury Mode Always ON except Always ON at home N/A system. Location history also shows that about 6 hours are
(user is sleeping
energy
spent in office and almost half of the time the desktop is used
550w*24*2/3= 50w*24*2/3=
insensitive) 8.8kWh 0.8kWh and for the other half time the laptop is used. The real total
Moderate Only ON when at Only ON when at energy consumption at office is 2.26kWh, which includes
Mode home awake Constantly, Constantly, home awake
185w/2*24 1.3kw*0.05=
1.15kWh for lighting, 0.96kWh for desktop, and 0.15kWh for
550w*24*1/3= 4.4 50w*24*1/3= 0.4
kWh = 2.22 kWh 0.065 kWh kWh laptop. For the remaining 4 hours, the user is not at home/office,
Frugal Mode Only 60% ON when Only 60% ON when and all the devices are in "OFF" status, except the refrigerator at
(user is at home awake at home awake
energy home.
sensitive)
4.4 *0.6 =2.64kWh Thus, for comparison, we put the real recorded energy
0.4*0.6 = 0.24kWh
consumption data after applying our ideas together with the
Total Luxury: 11.90kWh ; Moderate: 7.09kWh; Frugal: 5.17kWh
energy consumption estimation results of the three modes, to
*Assuming 8 hours working in office, 8 hours at home awake, and 8 hours sleeping
demonstrate how much energy can be saved. The results are
Then we apply our location based solution and dynamically
shown in Fig. 17. The simple takeaway message is that the real
control the appliances in both home and office to reduce the
energy consumption of the prototype system after applying our
energy waste and maximize the energy efficiency. We track
location based idea is very close to the frugal mode's energy
and record the location of the user in 24 hours' period and apply
consumption. It means that with our new idea, general users
dynamic control and policy changes in both home and office.
will enjoy luxury living style without special care or changes
The location history shown in Google map is in Fig. 16.
and they will pay what frugal mode users pay.
Note that for the detailed turning on/off policy changes, we
consider some real-life limitations. For example, in our testbed, C. System Implementation and Integration Challenges
we did not control the on/off status of the refrigerator. We only During the system prototyping process, we met multiple
apply changes to those devices such as lighting bulbs, desktop, challenges and found corresponding solutions to address these

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
9

issues. We discuss them here as follows. Therefore, to overcome such difficulty, we tuned the
1) Extending Control Device Functions threshold distance and did comprehensive experiments before
The first challenge for our system implementation was that reaching a threshold that could minimize the false alarm or
we needed to find appropriate devices and methods to carry out false positive rates. Our results proved to be satisfactory.
the control functions. More importantly, these methods should
not be limited by a specific software or platform. It is desirable
to allow the devices to be controllable from anywhere at any
time through the general PC platforms while not limited by
those dedicated platforms such as iOS or Android. Therefore,
we "hacked" the WeMo devices by wrapping up and extending
some standard uPnP protocol control module and successfully
realized the above functions.
20
18
Daily Energy (kWh)

16
5.78 Office
14
Home
12 Fig. 18. GPS location trace with an unmoved mobile user
10
8 3.02 D. Multi-scale Energy Proportionality
6 11.9 2.31 2.26 To summarize the experiments and results, we can see that
4 7.09 the effect of energy saving is conspicuous, although given
2 5.17 5.29 different types of buildings and occupants' energy using habits,
0 there can be different degrees of savings.
Luxury Moderate Frugal Real The above prototype system experiments vividly illustrate
Fig. 17. Comparison of the real energy consumption after applying our idea
with the three modes' energy estimation how user-scale energy proportionality is realized by using
networking and computing technologies, since the user's
2) Integrating Multiple Components energy consumption becomes approximately proportional to
As shown in Fig. 15, in our prototype, we have a series of his/her actual usage. Besides user-level energy proportionality,
sub-components to be integrated into the prototype system. For applying similar idea into multi-scale organizations in a
example, the client mobile devices with GPS sensor, the client building, as described in Section III, it virtually enables both
data uploading module, the data and web server, the controller, organization-level energy proportionality and building-level
the appliances under control, and the WiFi router with port energy proportionality. Specifically, when a particular user
mapping function. It has been a challenge to organize and controls and adjusts the appliance policies under his/her
integrate these heterogeneous devices to work together as a territory, he/she has to follow specialized group policies
coherent system. In our prototype, we successfully addressed enforced by the organization such as a laboratory or a
this challenge and implemented a working system. department. The organization also enforces the policies for
3) GPS Data Fast Parsing publicly shared parts such as HVAC, lighting, fire and safety,
In our prototype system, we used a GlobalSat GPS sensor, and elevator systems. It could designate special working staff to
and the coordinate data were encoded in NMEA-0183 data control and apply energy proportionality for these subsystems.
format. The GPS sensor generates a large amount of coordinate The laboratory or department aggregates each user' energy
data in real time which requires the server to receive them and proportionality and the publicly shared subsystems' energy
perform fast parsing to get the accurate coordinates of the proportionality, to achieve an organization scale energy
mobile user. We implemented the function in Python to fulfill proportionality. Similarly, the idea can be generalized to
this task, which also includes function implementing the building level since it basically aggregates multiple
distance calculation and the threshold comparison before organizations inside the building and multiple public
triggering the energy policy changes in both sides of the subsystems working for all the organizations in the building.
buildings in the testbed. Achieving multi-scale energy proportionality has profound
economic impacts to the society in terms of avoiding huge
4) Handling GPS Location Inaccuracy energy waste and saving costs for users and organizations. The
Another challenge was that the GPS coordinate data benefits would make a huge difference if the idea gets broadly
generated by the GlobalSat sensor has its own accuracy limit, implemented and deployed. The networking and computing
which means that even the mobile user does not move, the technologies used for the system enable the buildings running
generated coordinate data may vary. The following Fig. 18 and operation to be more intelligent and efficient, and in an
approximately demonstrates this effect. In the figure, we can automated manner without manual intervention, which is also
see that even for an unmoved mobile user, its coordinate data very important goal for future smart home/office applications.
trace may vary by 20 meters given the inaccuracy of the GPS All the stakeholders including the common occupants and users,
sensors. organizations' authorities, and buildings' owners and tenants

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
10

could have total flexible control over their own energy policies, VI. CONCLUSIONS
which is a very promising feature for our proposed idea. In this paper, we added new contributions besides
Moreover, the proposed system also involves every user and summarizing our previous work regarding the IoT framework
organization to participate in the energy saving efforts, which is for smart energy in buildings. The work includes: (1) energy
potentially a very good training and education method to consumption data analysis of the green building testbed, (2)
encourage everyone to study and participate in resolving global new smart location-based automated energy control framework
climate and sustainability issues in everyday life. Further designs, and (3) experimental prototype that applies IoT
incorporation of social network plugin into the smart mobile networking and computing technologies to improve the energy
phone based energy control platform would generate even efficiency in buildings. We put them into a complete three-step
broader impacts [2]. research and added significant new contributions proving the
ideas and concepts we proposed. By building this IoT
V. RELATED WORK framework in smart homes or offices, we aim to enable not only
multi-scale energy proportionality, but also create an intelligent
Due to the multi-disciplinary essence of the research topic,
home space which is an important part of the future smart world.
the related work covers a range of different areas. We discuss a We envision that the idea will provide not only significant
few of them briefly. Limited by the space, a longer list of the economic benefits but also huge social benefits in terms of
related work can be found in papers [2, 3]. global sustainability.
First related area is the building energy simulation. Many
building simulation software take building parameters as input VII. REFERENCES
and after processing, output estimated energy usage [16]. A
[1] USGBC Research Committee, “A National Green Building Research
popular one is “EnergyPlus” [12] provided by the Department Agenda,” November, 2007.
of Energy (DOE). In comparison, monitoring building energy [2] J. Pan, R. Jain, P. Biswas, et al., “A Framework for Smart Location-Based
consumption with real network system and analyzing the real Automated Energy Controls in a Green Building Testbed,” in the
energy consumption data can be more effective. Proceedings of IEEE EnergyTech 2012, Cleveland, OH, May 29-31,
2012.
Second area is the climate effect models research in which a [3] J. Pan, R. Jain, P. Biswas, et al., “Toward an Energy-Proportional
lot of existing work is about the relationship between energy Building Prospect: Evaluation and Analysis of the Energy Consumption
consumption and climate or weather factors [17, 18]. The in a Green Building Testbed,” in the Proceedings of IEEE Energytech
2013, Cleveland, OH, May 21-23, 2013.
related research consists of: (1) simulating the heat transfer [4] USGBC Leadership in Energy and Environmental Design (LEED), (2015,
processes and building structures (envelope, tree shelter, etc.) January 26). [Online]. Available:
to find how the climate can impact building energy efficiency; http://www.usgbc.org/DisplayPage.aspx?CategoryID=19
[5] U. S. Green Building Council, (2015, January 26). [Online]. Available:
(2) study of solar effects on heat and mass transfer and their http://www.usgbc.org/
impacts. Moreover, most of them are currently based on [6] R. Jain, “The Art of Computer Systems Performance Analysis:
theoretical thermal calculations and simulations and very few Techniques for Experimental Design, Measurement, Simulation, and
Modeling,” Wiley, 1991.
of them are using the actual building energy data and research
[7] Internet of Things, (2015, January 26). [Online]. Available:
on how to reduce the energy consumption by incorporating http://en.wikipedia.org/wiki/Internet_of_Things
occupants’ participation. A longer reference list of related [8] J. Pan, S. Paul, R. Jain, et al., “MILSA: A Mobility and Multihoming
efforts can be found at [19]. Supporting Identifier Locator Split Architecture for Next Generation
Internet”, IEEE Globecom 2008, New Orleans, LA, December 2008.
Third area is the application of Wireless Sensor Network [9] J. Pan, R. Jain, S. Paul, and C. So-In, “MILSA: A New Evolutionary
(WSN) technologies into the building environment and Architecture for Scalability, Mobility, and Multihoming in the Future
experiments on a specific subsystem like lighting and Internet,” IEEE Journal on Selected Areas in Communications (JSAC),
Special issue on Routing Scalability, October 2010.
thermostat. For example, it is used to sense and control the [10] S. Paul, R. Jain, J. Pan, M. Bowman, "A Vision of the Next Generation
lights according to the detection results of the sunlight for a Internet: A Policy Oriented Perspective," British Computer Society
building based on human activities, to monitor the electrical (BCS) International Conference on Visions of Computer Science,
Imperial College, London, September 22-24, 2008.
energy consumption, to log the human activities and to adjust [11] R. Jain, “Internet 3.0: Ten Problems with Current Internet Architecture
the HVAC working time to provide better comfort, etc. There and Solutions for the Next Generation,” IEEE MILCOM 2006,
are already a lot of such experimental researches. Typical ones Washington, DC, October 23-25, 2006.
[12] EnergyPlus Energy Simulation Software, U.S. Department of Energy,
are in [20, 21] and more can be found from the list at [19]. (2015, January 26). [Online]. Available:
However, WSN were originally designed for other purposes but http://apps1.eere.energy.gov/buildings/energyplus/
they are able to provide a good complement to other sensing [13] Cloud Computing, (2015, January 26). [Online]. Available:
http://en.wikipedia.org/wiki/Cloud_computing
and metering technologies in the building environment.
[14] P3 International, Kill-A-Watt product series,
Last related category lies in the information and computer http://www.p3international.com/products/p4400.html
science technologies. The iPhone, Android, and Windows [15] Belkin WeMo switch, (2015, January 26). [Online]. Available:
Phones provide similar open platforms to develop versatile http://www.belkin.com/us/wemo
[16] Building Energy Software Tools Directory, U.S. Department of Energy,
smart applications with multiple sensors including GPS. So far, (2015, January 26). [Online]. Available:
there are not many wide-scale applications on building http://apps1.eere.energy.gov/buildings/tools_directory/.
environment energy auditing and control. Cloud computing [13] [17] F. Flouquet, “Local weather correlations and bias in building parameter
estimates from energy-signature models,” Energy and Buildings, Volume
certainly is a very hot topic related to our research. 19, Issue 2, Pages 113-123, 1992.

U.S. Government work not protected by U.S. copyright.


This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JIOT.2015.2413397, IEEE Internet of Things Journal
11

[18] K. Wan, D. Li, J. Lam, "Assessment of climate change impact on building which won the 1991 Best-Advanced How-to Book, Systems Award from the
energy use and mitigation measures in subtropical climates," Energy, Computer Press Association.
Volume 36, Issue 3, Pages 1404-1414, March, 2011.
[19] Reference List for Building Energy Modeling, (2015, January 26). Subharthi Paul [M] is currently a member of technical staff in Cisco, INC. He
[Online]. Available: http://cse.wustl.edu/~jp10/energyref.htm received his Ph.D. from Washington University in Saint Louis, MO USA.
[20] X. Jiang, M. V. Ly, J. Taneja, “Experiences with a High-Fidelity Wireless Before that, he received his B.S. degree from the University of Delhi, India, and
Building Energy Auditing Network,” Proc. of the 7th ACM Conference his Master’s degree in software engineering from Jadavpur University,
on Embedded Networked Sensor Systems (SenSys 2009), New York, Kolkata, India. His primary research interests are in the area of future Internet
2009. architectures.
[21] V. Singhvi, A. Krause, C. Guestrin, “Intelligent Light Control using
Sensor Networks,” Proc. of the 3rd international conference on
Tam Vu [M] is currently an assistant professor in the Department of Computer
Embedded networked sensor systems (SenSys 2005), 2005.
Science and Engineering at the University of Colorado, Denver. He holds a
Ph.D. from the Department of Computer Science at Rutgers University. He got
the Best Paper Awards at the Annual International Conference on Mobile
Jianli Pan [M] is currently an assistant professor of Computer Science in the
Computing and Networking (MobiCom) in 2011 and 2012. His current
Department of Mathematics and Computer Science at the University of
research is on mobile systems and wireless networks and their various
Missouri - St. Louis, MO USA. Previously he obtained his Ph.D. degree in
properties.
Computer Engineering from Washington University in Saint Louis, MO USA.
He also holds two master degrees from Washington University in Saint Louis
and Beijing University of Posts and Telecommunications, respectively. His Abusayeed Saifullah [M] is an assistant professor in the Computer Science
current research is on the Future Internet Architecture, mobile and cloud Department at Missouri University of Science & Technology. He received PhD
computing, network virtualization, Internet of Things, cyber security, and smart in 2014 from the Computer Science and Engineering Department at
energy in buildings and smart grids. Washington University in St Louis. His research primarily concerns
cyber-physical systems with contributions spanning real-time systems,
embedded systems, wireless sensor networks, and parallel and distributed
Raj Jain [F] is a Fellow of ACM and AAAS, a winner of the ACM SIGCOMM
computing. He received the Best Paper Award at RTSS ’14, the Best Student
Test of Time Award and CDAC-ACCS Foundation Award 2009, and ranks
Paper Award at RTSS ’11, Best Paper Nomination at RTAS’ 12, and the Best
among the top 100 in CiteseerX’s list of Most Cited Authors in Computer
Student Paper Award at ISPA ’07.
Science. He is currently a professor in the Department of Computer Science and
Engineering at Washington University. Previously, he was one of the
cofounders of Nayna Networks, Inc., a next-generation telecommunications Mo Sha [M] is currently an assistant professor in the division of science of the
systems company in San Jose, California. He was a senior consulting engineer Governors State University, Illinois. He obtained his Ph.D. degree from
at Digital Equipment Corporation in Littleton, Massachusetts, and then a Washington University in Saint Louis.
professor of computer and information sciences at Ohio State University,
Columbus. He is the author of Art of Computer Systems Performance Analysis,

U.S. Government work not protected by U.S. copyright.

También podría gustarte